Study On The Recognition Of Typical Soil In Inner Mongolia Grassland Based On Hyperspectral

Grassland soil is an important part of grassland ecosystem.In this research,I select several typical grasslands in Inner Mongolia Autonomous Region to collect hyperspectral ground image data under natural light irradiation,and use spectrum and convolution neural network to identify and classify the collected image data,and achieve high accuracy.In this paper,the grassland covered soil is taken as the research object.Through the minimum noise separation of hyperspectral data of several soils,the bands with high contribution rate are selected,and the spectral analysis is carried out by differential method and developing method.In addition,in view of the characteristics of high dimension and high redundancy of hyperspectral image,PCA method is used to reduce the dimension,and the data after dimension reduction is input into the improved lenet5 convolutional neural network,and the most suitable reduced dimension waveband number and learning rate are selected for training in many experiments.Finally,the testing accuracy of the Convolutional neural network model reaches 94.7%,which meets the classification accuracy of grassland soil hyperspectral image At the same time,compared with the typical SVM method.In this paper,the method of grassland soil classification in Inner Mongolia can provide a method and basis for the dynamic monitoring of grassland soil degradation by remote sensing.